TIDES: Implicit Time-Awareness in Selective State Space Models
TIDES keeps physical time gaps in selective state-space models and tops irregular time-series benchmarks.
TIDES is a selective state-space model that keeps the discretization step equal to the physical time gap while moving input dependence onto the diagonal state matrix. That design handles irregular timestamps like continuous-time SSMs such as S5 without losing Mamba-style per-token expressivity. On a Fading Flash diagnostic it avoids failure modes of current architectures. TIDES sets the best average rank on UEA time-series classification and Physiome ODE regression, and matches or exceeds the reference baseline on 6 of 8 natively irregular datasets.
- TIDES puts input dependence on the diagonal state matrix, not the step size.
- The discretization step stays the physical time gap, supporting irregular timestamps.
- It sets the best average rank on UEA classification and Physiome ODE regression.
- It matches or beats the reference baseline on 6 of 8 irregular datasets.
From the original article
1 min read at huggingface.co
Selective state space models (SSMs), such as Mamba, achieve strong per-token expressivity by making the time discretization step TildeΔ a learned function of the input. However, in doing so, TildeΔ no longer equals the physical time gap Δ between consecutive observations, limiting the ability of these models to handle irregular time series.
By Taylan Soydan, Miguel A. Bessa, Dirk Mohr · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.